Do Emotions Influence Moral Judgment in Large Language Models?

2026-04-21

AI Summary

This study investigates how emotions systematically influence the moral judgments of large language models (LLMs), addressing a critical gap at the intersection of affective computing and machine moral reasoning. By constructing an emotion induction pipeline that injects specific emotional states into diverse moral scenarios, the authors employ multi-dataset evaluation, model capability scaling analysis, and human对照 experiments to demonstrate—for the first time—that emotions exert directional effects on LLMs’ moral acceptability judgments: positive emotions elevate and negative emotions diminish judgment scores, with binary moral conclusions reversed in approximately 20% of cases. The research further reveals that higher-capability models exhibit reduced emotional sensitivity, and certain emotions—such as regret—produce effects contrary to valence-based expectations, highlighting misalignments between model and human moral reasoning.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Affective Computing
• Humans and AI: Emotional Intelligence

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Large language models have been extensively studied for emotion recognition and moral reasoning as distinct capabilities, yet the extent to which emotions influence moral judgment remains underexplored. In this work, we develop an emotion-induction pipeline that infuses emotion into moral situations and evaluate shifts in moral acceptability across multiple datasets and LLMs. We observe a directional pattern: positive emotions increase moral acceptability and negative emotions decrease it, with effects strong enough to reverse binary moral judgments in up to 20% of cases, and with susceptibility scaling inversely with model capability. Our analysis further reveals that specific emotions can sometimes behave contrary to what their valence would predict (e.g., remorse paradoxically increases acceptability). A complementary human annotation study shows humans do not exhibit these systematic shifts, indicating an alignment gap in current LLMs.

Problem

• emotion
• moral judgment
• large language models
• alignment gap
• moral reasoning

Innovation

• emotion induction
• moral judgment
• large language models
• alignment gap
• valence effects